Academia.eduAcademia.edu

Figure 8 – uploaded by hadis anahideh

See full PDFdownloadDownload figure
ness metrics, Figure 8, we illustrate a graph with nodes rep- resenting the fairness metrics, orange nodes indicating the representative metrics of each subset, and edges showing the subsets of metrics that are highly correlated. We used T = 0.5 for this plot. The graph confirms that the number of subsets are smaller using Logit model than that using NN, as previously discussed. Similarly, Figure 9 shows the rep- resentative metrics for Credit dataset using Logit and NN models. Discovery Ratio, Predictive Parity, Equality of Op- portunity, and Average Odd Difference are selected as repre- sentative metrics for Credit dataset when we use Logit clas- sifier.

Figure 8 ness metrics, Figure 8, we illustrate a graph with nodes rep- resenting the fairness metrics, orange nodes indicating the representative metrics of each subset, and edges showing the subsets of metrics that are highly correlated. We used T = 0.5 for this plot. The graph confirms that the number of subsets are smaller using Logit model than that using NN, as previously discussed. Similarly, Figure 9 shows the rep- resentative metrics for Credit dataset using Logit and NN models. Discovery Ratio, Predictive Parity, Equality of Op- portunity, and Average Odd Difference are selected as repre- sentative metrics for Credit dataset when we use Logit clas- sifier.

Related Figures (15)

randomness over fairness values space. Besides, since the correlation estimator calls the oracle many times before it computes the correlations, we want the oracle to be efficient.
randomness over fairness values space. Besides, since the correlation estimator calls the oracle many times before it computes the correlations, we want the oracle to be efficient.
Figure |: Table of fairness values for the NV sampled models Groun 1 Group 2 ay rR
Figure |: Table of fairness values for the NV sampled models Groun 1 Group 2 ay rR
Figure 4: Using COMPAS data set to illustrate that Correlations are model dependent.
Figure 4: Using COMPAS data set to illustrate that Correlations are model dependent.
Figure 5: Using Logistic Regression to illustrate that Correlations are data dependent. Figure 7: Number of representative subsets given a model using different dataset
Figure 5: Using Logistic Regression to illustrate that Correlations are data dependent. Figure 7: Number of representative subsets given a model using different dataset
Algorithm 2: SamplingOracle
Algorithm 2: SamplingOracle
(a) Correlation Estimate for f7 and fi2 for Credit- age
(a) Correlation Estimate for f7 and fi2 for Credit- age
Figure 13 illustrate the correlation estimations for Credit- age dataset. Note that in (c) the correlation estimates for certain pairs are missing, when we use NN. The reason for missing correlations is the NA values for fg which indicate that F'N; is zero for this case. Similarly for f14, since F'Ny is zero it yiels NA for FNR ratio which is f;4. In addition, fg is always zero for any sampled model (FOR disparity rate is zero). Similarly for {13 the disparity is zero.
Figure 13 illustrate the correlation estimations for Credit- age dataset. Note that in (c) the correlation estimates for certain pairs are missing, when we use NN. The reason for missing correlations is the NA values for fg which indicate that F'N; is zero for this case. Similarly for f14, since F'Ny is zero it yiels NA for FNR ratio which is f;4. In addition, fg is always zero for any sampled model (FOR disparity rate is zero). Similarly for {13 the disparity is zero.
Figure 11: Number of representative subsets given a model using different datasets (Logit, RF, KNN)
Figure 11: Number of representative subsets given a model using different datasets (Logit, RF, KNN)
Figure 12: Number of representative subsets given a model using different datasets (SVM, NN)
Figure 12: Number of representative subsets given a model using different datasets (SVM, NN)
(a) Correlation values for fg and f7(b) Correlation values for fz and fg(c) Correlation values for {11 and(d) Correlation values for fs and for Logit for Logit fis for Logit fic for Logit Figure 15: Correlation Estimates for different pairs on COMPAS dataset using different models.
(a) Correlation values for fg and f7(b) Correlation values for fz and fg(c) Correlation values for {11 and(d) Correlation values for fs and for Logit for Logit fis for Logit fic for Logit Figure 15: Correlation Estimates for different pairs on COMPAS dataset using different models.
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved